autoresearch — Autonomous Improvement Loop

Teach your agent to improve itself (or your code, prompts, configs) using the pattern open-sourced by Andrej Karpathy. The agent modifies one file, measures one metric, and keeps only what improves the score. Repeat until you wake up to a better system.

"You're not touching any of the Python files. Instead, you are programming the program.md Markdown files that provide context to the AI agents." — Karpathy

The Pattern (3 Constraints)

Before running any loop, define these three things:

FILE_TO_MODIFY = the ONE file the agent can edit
METRIC         = the ONE number that defines "better"
BUDGET         = the fixed time/compute per experiment

Then loop: